Good article count. Focus on topic coverage depth, not just volume.
Content is current. Set a quarterly review calendar to maintain it.
Neither search nor structure — customers can't find answers even if they exist.
No signal on what's missing. Add ratings and weekly unanswered-question review.
Coverage and freshness drive 80% of deflection quality.
A knowledge base scores well when it covers the main question categories, stays current, and gives customers a way to find answers. Coverage and freshness together determine whether the AI has accurate source material. A large library of outdated articles scores worse than a small, accurate one.
Findability and feedback loop are multipliers. Good content that is hard to search or navigate gets retrieved less reliably by an AI agent. And without a weekly review of unanswered questions, coverage gaps compound silently until they show up as escalations.
Sources & methodology3 sources
KB coverage is positively correlated with self-service deflection rate. Organizations with 50+ articles see significantly higher first-contact resolution versus those with fewer than 20.
Content older than 12 months shows a measurably higher error rate; the freshness scoring tiers in this tool reflect Gartner's content decay curve research.
Teams that review unanswered questions weekly see 2× faster knowledge base improvement than those on a monthly review cycle, used to weight the feedback loop dimension.
Frequently asked questions
- What is a good KB health score?
- 70 or above (B grade) means your knowledge base is functioning well and can support meaningful AI deflection. Below 50 means the coverage, freshness, or findability gaps will limit how much an AI agent can help — those gaps produce unanswered questions, not deflected tickets.
- Which dimension has the biggest impact on AI deflection?
- Coverage first, then freshness. An AI agent can only answer what exists in your knowledge base. A well-maintained library of 50+ articles covering your core question categories typically unlocks 40–60% deflection. Findability amplifies coverage — good content that isn't structured well gets retrieved less reliably.
- How often should I audit my knowledge base?
- A lightweight weekly review of unanswered questions (what did the AI fail to answer this week?) is more effective than a big monthly audit. The weekly signal tells you exactly what to write next. A full audit of existing content freshness makes sense quarterly, or whenever your product ships a significant update.